Claude Opus 4.8 Is Out Today — Same Price, Bigger Gains

Anthropic just dropped Claude Opus 4.8 this morning — May 28, 2026 — less than six weeks after Opus 4.7 launched in April. Same price. Meaningfully better across the board.

Anthropic describes Opus 4.8 as having "sharper judgement, more honesty about its progress, and the ability to work independently for longer than its predecessors." Those aren't hollow release-day adjectives. The benchmark numbers back it up.

The Benchmark Numbers

Agentic coding (SWE-Bench Pro) jumps from 64.3% to 69.2%. Multidisciplinary reasoning with tools goes from 54.7% to 57.9%. Agentic computer use ticks up from 82.8% to 83.4%. Knowledge work score moves from 1,753 to 1,890.

The coding number is the one that matters most. At 69.2% on SWE-Bench Pro, Opus 4.8 is 10 points ahead of GPT-5.5.

On Anthropic's own Super-Agent benchmark, Opus 4.8 is the only model to complete every case end-to-end, beating prior Opus models and GPT-5.5 at parity on cost.

Dynamic Workflows: The Headline Feature

This is what developers should pay the most attention to.

Dynamic Workflows is a new research preview feature in Claude Code that allows Claude to take on even bigger tasks. Claude can plan the work and then run hundreds of parallel subagents in a single session — with Opus 4.8, the agents can run for even longer. It then verifies its outputs before reporting back to the user.

In practice: you hand Claude Code a massive task — full codebase refactor, multi-service migration, large-scale test generation — and it breaks it into parallel workstreams, runs them simultaneously, then audits the results before surfacing anything to you. That's a fundamentally different capability from a single sequential agent loop.

Effort Control: You Decide How Hard It Thinks

Users on claude.ai and Cowork can now select how much thinking effort Claude applies — from Low (faster, lower rate-limit usage) to Max. Opus 4.8 defaults to High effort for the best balance of quality and experience.

Practically: Low for quick questions and drafts. High for complex tasks. Max when you need every last drop of reasoning and don't care how long it takes. This also means you can stretch your rate limits further on tasks that don't need full horsepower.

Fast Mode Is Now 3x Cheaper

Fast mode for Opus 4.8 — where the model works at 2.5x the speed — is now three times cheaper than it was for previous models.

Developers can activate it directly in Claude Code using the /fast command. API access is available by contacting an account manager or joining the waitlist.

That price drop is significant. Fast mode was previously expensive enough that it made more sense to just use a smaller model for latency-sensitive tasks. That calculus has shifted.

Honesty and Reliability Improvements

The model is four times less likely than Opus 4.7 to miss flaws in code it produces and is less prone to unsupported claims. Early testers report that Opus 4.8 is more likely to flag uncertainties about its work.

Less hallucination, more "I'm not sure about this." For long-running agentic tasks especially, a model that flags its own uncertainty is genuinely more useful than one that confidently produces plausible-but-wrong outputs at the end of a 40-minute session.

Pricing and Access

Opus 4.8 is available today at the same price as Opus 4.7: $5 per million input tokens and $25 per million output tokens, with up to 90% cost savings with prompt caching and 50% savings with batch processing. Use claude-opus-4-8 via the Claude API.

For Claude.ai users, Opus 4.8 is available on Pro, Max, Team, and Enterprise plans. On the API side, it's available natively and through AWS, Google Cloud, and Microsoft Foundry.

What's Coming Next

Anthropic noted in the release that they plan to release "a new class of model with even higher intelligence than Opus." Mythos-class models are expected in the coming weeks — with mid-June the current estimate circulating in the developer community.

Opus 4.8 is the last stop before something significantly bigger.

Sources: Anthropic, 9to5Mac, OfficeChai, Axios, AWS